Full-stack AI, the transoceanic resonance between Alibaba and Google

In the fiercely turbulent global AI race of 2025, Alibaba and Google, two leading global AI companies, are coincidentally building a full-stack AI ecosystem that connects the complete closed loop from underlying hardware to applications, becoming the only two top technology companies in the world with full-stack AI technology capabilities.

The cornerstone of this offensive is its unprecedented commitment to capital expenditure.

Today, Alibaba's Q1 fiscal year 2026 (natural year Q2 2025) financial report revealed a quarterly capital expenditure (Capex) of 38.6 billion yuan, setting a new historical high. In the earnings call, CEO Eddie Wu also announced that over the past four quarters, Alibaba has cumulatively invested more than 100 billion yuan in AI infrastructure and AI product research and development.

The effectiveness of AI investments is evident, with strong performance in AI + cloud this quarter, as Alibaba Cloud's revenue growth continues to accelerate to 26%, and AI-related product revenue has achieved triple-digit year-on-year growth for eight consecutive quarters. Additionally, Alibaba disclosed for the first time that the proportion of quarterly AI revenue in external commercialization revenue has exceeded 20%.

Supported by the depth and intensity of AI investments, Alibaba's large model innovation engine is operating at an astonishing pace. In July, Alibaba successively released and open-sourced several heavyweight models: the Qwen3 inference model, which topped the world's strongest open-source inference models; the new version of the open-sourced Qwen3, the strongest in classic foundational models; the Qwen3-Coder, which has programming capabilities comparable to the world's top closed-source models; and the industry's first video generation model Wan2.2 using MoE architecture released at the end of July.

In August, Alibaba open-sourced the new text-to-image model Qwen-Image, which immediately topped the Hugging Face model leaderboard. On August 22, Alibaba also launched the Agentic programming platform Qoder.

This high-frequency model iteration is rapidly building an ecological moat through an open-source strategy. Currently, the number of derivative models from Tongyi Qianwen has surpassed 140,000, surpassing Llama to become the world's largest AI open-source model, with global downloads exceeding 400 million.

This series of actions perfectly illustrates the core strategy of "user-first, AI-driven" established by Eddie Wu.

Transpacific Resonance: Google's Heavy Bet on AI and the Emergence of the Full-Stack Model

Alibaba's AI strategic ambitions have received strong resonance across the Pacific. Google's Q2 2025 financial report clearly brought the "full-stack AI" strategic model to the forefront.

The most striking data in the report is that Google has doubled its token usage in a month by enhancing model capabilities and offering various applications at low or no cost, jumping from 480 trillion in May to 980 trillion.

To cope with the explosive demand for inference, Google has significantly raised its capital expenditure (Capex) guidance for fiscal year 2025 from $75 billion to an astonishing $85 billion. This gamble, described by Wall Street analysts as "un-Googly," is in stark contrast to its past "prudent and mature" investment style Despite the fact that this move has severely compressed the company's free cash flow (FCF) in the short term, analysts generally believe that this is precisely what investors have been "strongly calling for," indicating that Google is finally ready to "take AI seriously."

Google's "irrational" gamble is aimed at building and strengthening its full-stack AI capabilities.

The core of this model lies in the fact that tech giants must control every key link from the bottom to the top to achieve extreme optimization of performance, cost, and innovation speed.

This model can be clearly divided into three levels:

1. Hardware/Infrastructural Layer: Including self-developed chips, server clusters, and a global network of data centers.
2. Base Model Layer: Possessing a powerful proprietary large model that serves as the "intelligent brain," such as Google's Gemini.
3. Application/Ecosystem Layer: Relying on market-dominating applications (such as Google Search and YouTube) as the main channels for AI technology deployment, data collection, and user reach.

The fact that Google and Alibaba have both chosen this capital-intensive, full-stack technology path is no coincidence.

This proves that the global AI competition paradigm is changing: the era of pure algorithm competition has ended, replaced by a comprehensive war over integrated systems, capital strength, and ecosystem control. This full-stack AI model is becoming the only path for companies aspiring to achieve AI leadership in global or regional markets, significantly raising the industry's entry barriers.

Full-Stack AI Layout: Hardware, Models, and Applications Advancing Together

Through an in-depth analysis of Alibaba and Google's AI strategies, it can be observed that both exhibit a striking "strategic mirror" across the three levels of full-stack layout. Although specific tactics may differ, their underlying logic is highly consistent.

1. Computing Power Foundation: Building a Trillion-Level AI Infrastructure Platform

Controlling physical infrastructure is the starting point of the full-stack strategy. This concerns the performance and cost of the entire downstream ecosystem, and both Alibaba and Google regard owning and operating their own computing power as an unshakeable cornerstone.

Alibaba plans to invest 380 billion RMB over the next three years, with a clear goal of "building a cloud computing network with global technological competitiveness."

Since the beginning of this year, Alibaba Cloud has launched eight new AI and cloud data centers and availability zones globally, including in Beijing, Shanghai, Hangzhou, as well as Thailand, South Korea, Malaysia, Dubai, and Mexico, marking the initial results of this plan. In the second half of this year, Alibaba Cloud's global infrastructure layout will expand to 30 regions and 95 availability zones.

Google's $85 billion capital expenditure plan also focuses on this, with about two-thirds allocated for servers and one-third for data center construction and network equipment, primarily aimed at meeting the strong demand for AI computing power from cloud customers and supporting its vast internal AI services 2. Model Hub: Tongyi Qwen and Gemini as the Ecological Brain

If computing power is the "body," then large models are the "brain." Both Alibaba and Google position their flagship models as the core hub for building developer ecosystems, rather than merely as a product.

Alibaba's Tongyi Qwen series has chosen an open-source-led ecological path. With its advantages of high performance and low cost, Qwen has rapidly ascended to the throne of the world's strongest open-source model this year, with over 140,000 derivative models globally and downloads exceeding 400 million.

Through China's largest AI open-source community, "ModelScope," Alibaba has gathered over 16 million developers to jointly build a vast AI ecosystem centered around Tongyi.

Google's Gemini, on the other hand, has taken a different route—deep integration and large-scale application. Through its extensive existing developer network, Gemini has attracted over 9 million developers to build. Its processing capacity has reached an astonishing monthly total of over 980 trillion tokens, doubling since May. This clearly demonstrates Google's powerful capability to deeply embed Gemini into the global developer workflow by leveraging its platform advantages.

3. Application is King: From E-commerce and Search to AI Implementation in Various Scenarios

The ultimate value of a full-stack model is reflected at the application layer.

Both Alibaba and Google cleverly utilize their absolute advantages in C-end applications as the best testing ground, distribution channel, and data source for AI technology, forming a powerful positive feedback loop.

Alibaba's flywheel is driven by its vast commercial ecosystem. Recently, Amap has been fully AI-enabled, DingTalk has completed its latest AI upgrade, and the Taobao platform is upgrading a series of AI applications such as AI search and AI advertising platforms. Financial reports show that Taotian's AI tool "Quanzhan Tui" continues to enhance the operational efficiency of Taotian merchants. In June, Taobao launched the billion-parameter large model RecGPT, making "You May Also Like" more accurate, with tests showing that user add-to-cart frequency and time spent increased by over 5%. Alibaba International's AI tool Marco has an average daily call volume exceeding 1 billion times.

Google's flywheel revolves around its core search and cloud business. The AI Overviews feature in search has reached over 2 billion monthly active users and increased query volume for related searches by 10%. On the enterprise side, Google Cloud has effectively driven a 32% revenue growth by bundling Gemini features into the Workspace suite.

The logic of this closed-loop system is clear: the application layer (such as e-commerce and search) generates massive amounts of high-quality proprietary data and user feedback, which are used to train and optimize the model layer (Tongyi, Gemini); while stronger models, in turn, enhance the application experience, attracting more users and generating more data.

The entire cycle operates efficiently on its own optimized infrastructure layer, forming an insurmountable competitive barrier.

Alibaba's full-stack AI advantage perfectly combines with a complete, end-to-end commercial ecosystem, constituting its core technological barrier—a highly efficient "AI flywheel."

Conclusion: It's Not Just a Technological Competition, But a Battle of Business Ecosystems

Alibaba and Google are converging in their AI strategies, both opting for a capital-intensive, technology-vertical full-stack model, which may become a key feature defining the landscape of the AI era.

This heralds the arrival of a new era: future competition will no longer be a contest of single algorithms or products, but a showdown of complete ecosystems ranging from chips, data centers, and large models to killer applications. This is not only a ticket to the top arena but also the ultimate weapon that determines victory or defeat.

For Alibaba, the significance of implementing a full-stack AI strategy is particularly profound. On a solid AI infrastructure platform, by injecting AI capabilities into various industries such as e-commerce, logistics, transportation, search, office, and finance, Alibaba is building a self-reinforcing, efficiently monetized business intelligence closed loop.

Looking at the Chinese market, the outcome of this AI competition will not depend on who has the smartest single model, but on who can build the most powerful, comprehensive, and synergistic AI-driven business ecosystem. In this ultimate showdown, Alibaba, which already possesses a complete business closed loop and is fully committed to full-stack AI, undoubtedly occupies a favorable strategic high ground